Flight paths of seabirds soaring over the ocean surfaceenable measurement of fine-scale wind speedand directionYoshinari Yoneharaa,1,2, Yusuke Gotoa,1, Ken Yodab, Yutaka Watanukic, Lindsay C. Youngd, Henri Weimerskirche,Charles-André Boste, and Katsufumi Satoa

aAtmosphere and Ocean Research Institute, The University of Tokyo, Kashiwa, Chiba 277-8564, Japan; bGraduate School of Environmental Studies, NagoyaUniversity, Furo, Chikusa, Nagoya 464-8601, Japan; cGraduate School of Fisheries Sciences, Hokkaido University, Minato, Hakodate 041-8611, Japan; dPacific RimConservation, Honolulu, HI 96822; and eCentre d’Etudes Biologiques de Chize (CEBC), UMR 7372 CNRS, Université de La Rochelle, 79360 Villiers-en-Bois, France

Edited by James A. Estes, University of California, Santa Cruz, CA, and approved June 10, 2016 (received for review December 3, 2015)

Ocean surface winds are an essential factor in understandingthe physical interactions between the atmosphere and the ocean.Surface winds measured by satellite scatterometers and buoyscover most of the global ocean; however, there are still spatialand temporal gaps and finer-scale variations of wind that may beoverlooked, particularly in coastal areas. Here, we show that flightpaths of soaring seabirds can be used to estimate fine-scale (every5 min, ∼5 km) ocean surface winds. Fine-scale global positioningsystem (GPS) positional data revealed that soaring seabirds flewtortuously and ground speed fluctuated presumably due to tailwinds and head winds. Taking advantage of the ground speeddifference in relation to flight direction, we reliably estimatedwind speed and direction experienced by the birds. These bird-based wind velocities were significantly correlated with wind ve-locities estimated by satellite-borne scatterometers. Furthermore,extensive travel distances and flight duration of the seabirds en-abled a wide range of high-resolution wind observations, espe-cially in coastal areas. Our study suggests that seabirds provide aplatform from which to measure ocean surface winds, potentiallycomplementing conventional wind measurements by covering spa-tial and temporal measurement gaps.

biologging | dynamic soaring | satellite scatterometer | GPS | meteorology

Recently, remote-sensing systems used to record atmosphericcirculation have been developed. Satellite-borne scatterometers

estimate ocean surface wind velocities each day covering themajority of the global ocean. These wide-range wind data incombination with refined ocean models are used in numericalweather predictions and describe the oceanographic featuresmore accurately (1–3). Buoys scattered over the ocean alsomeasure fine-time resolution in situ surface winds and are usedin validating remote-sensing measurements and are assimilatedinto model analyses (4, 5). However, because wind data are onlyacquired twice per day by each satellite and buoys have limitedspatial coverage, finer-scale changes of hours to days in localwind conditions could be overlooked. In addition, in coastalareas, where high biological productivity is associated with oceanand atmosphere circulation patterns (6), wind data are lackingdue to variations in wind and wave caused by complex topo-graphic effects that satellites have difficulty measuring (5, 7, 8).Obtaining in situ high-resolution atmospheric and oceanographicdata to fill these spatial and temporal observation gaps woulddeepen our understanding of physical processes relevant to in-teractions between the atmosphere and ocean, contribute toimproved atmospheric and ocean model analyses (7, 8), andreveal detailed structure that remains unresolved by usingconventional methods (9).The recent development of miniaturized animal-borne data

loggers presented a capability to use animals as indicators ofenvironmental variables. The extensive movement range andlocomotion ability of marine mammals and seabirds enable

observations to be obtained in places and scales unresolved byconventional observations. For example, instrumented seals havebeen providing temperature and salinity profiles in the AntarcticOcean for more than 10 y, especially under sea ice coverage thatwas difficult to measure by conventional methods (10, 11).Adding these data to ocean circulation models improved theaccuracy of estimated mixed layer properties (12). Bird-bornesensors are also used in measuring environmental variables suchas temperature, depth, and light intensity directly from the in-struments carried by the animal (13–16). Besides direct mea-surement from animal-borne instruments, indirect evaluation offlow velocity can be made particularly when animal movementsare passively driven or strongly affected by the flow. Studies haveevaluated velocity of air and water flows from bird movementtrajectories, which is the consequence of bird movement itselfand the drift caused by the flow. For example, wind velocity andthe state of a bird in relation to the wind can be evaluated usingcircular statistical models (17). Three-dimensional flight paths ofthermal soaring raptors have been used to estimate the hori-zontal and vertical component of wind in the mountain regionsand agreed with measurements from meteorological stations(18). Furthermore, movements of shearwaters floating on the

Significance

Monitoring ocean surface winds is essential for understandingocean and atmosphere interactions and weather forecasts.However, wind measured by satellite scatterometers and buoysare spatially and temporally coarse, particularly in coastal areas.We deployed small global positioning system units on soaringseabirds to record their tracks. Seabirds were accelerated by tailwinds or slowed down by head winds during flight, so theirflight speed changed in relation to wind speed and direction.Taking advantage of these changes in flight speed, we reliablyestimated wind speed and direction experienced by the seabirds.The wind observed by soaring seabird’s tracks complementedthe conventional observation gaps in terms of both time andspace, suggesting the possibility of using soaring seabirds asa living anemometer.

Author contributions: Y.Y., Y.G., K.Y., and K.S. designed research; Y.Y., Y.G., K.Y., Y.W.,L.C.Y., H.W., C.-A.B., and K.S. performed research; Y.Y. analyzed data; and Y.Y. wrotethe paper.

The authors declare no conflict of interest.

This article is a PNAS Direct Submission.

Freely available online through the PNAS open access option.

Data deposition: Time series of latitude and longitude data of all the studied birds havebeen deposited in the Dryad Digital Repository, datadryad.org (10.5061/dryad.3pb86).1Y.Y. and Y.G. contributed equally to this work.2To whom correspondence should be addressed. Email: yonehara@aori.u-tokyo.ac.jp.

This article contains supporting information online at www.pnas.org/lookup/suppl/doi:10.1073/pnas.1523853113/-/DCSupplemental.

www.pnas.org/cgi/doi/10.1073/pnas.1523853113 PNAS | August 9, 2016 | vol. 113 | no. 32 | 9039–9044

ENVIRONMEN

TAL

SCIENCE

S

Dow

nloa

ded

by g

uest

on

Oct

ober

5, 2

020

ocean surface were used to derive high-resolution ocean surfacecurrents that matched with in situ and remote-sensing mea-surements of currents (19). These seabird-derived current datawere assimilated into ocean models, resulting in refinement ofthe gyre patterns represented by the model (20). These studiesdemonstrate the potential of animal-derived environmental datain meteorological observations.Here, we propose a simple method to use soaring seabirds as

a bioindicator to estimate ocean surface winds. We deployedglobal positioning system (GPS) units on the backs of three speciesof soaring seabirds, streaked shearwater (Calonectris leucomelas;mean body mass, 0.6 kg), Laysan albatross (Phoebastria immutabilis;3.1 kg), and wandering albatross (Diomedea exulans; 9.7 kg) toinvestigate fine-scale flight trajectories by recording one positionper second. We further estimated wind velocities from the flighttrajectories of the birds, taking advantage of the ground speedchange caused by wind resistance and assistance. We examinedthe accuracy of the bird-based wind velocity estimations anddiscuss possible effects of a bird’s flight strategy to the windestimation.

Estimating Ocean Surface Wind Using Soaring SeabirdsWe obtained a total of 353, 74, and 185 h of positional data fromstreaked shearwaters (27.2 ± 12.9 h, n = 13), Laysan albatrosses(37.1 ± 4.0 h, n = 2), and wandering albatrosses (46.3 ± 1.2 h, n =4), respectively. Flight paths of shearwaters and albatrossesshowed a tortuous pattern in fine-scale movement on the orderof several tens of meters and ground speed fluctuated duringflight (Fig. 1). We assumed that the fine-scale ground speed ofsoaring seabirds was mainly affected by wind along the flightdirection of the bird (21, 22). According to this assumption, theground speed of a bird would be maximized in tail winds andequals the sum of air speed and wind speed, whereas it would beminimized in head winds and equals the wind speed subtractedfrom air speed. In side winds, ground speed increases as the tail

wind component increases. By fitting a sinusoidal curve to therelationship between ground speed and flight direction of each5-min (301 positional fix) section of flight (Fig. 1 C and F, andMaterials and Methods), we were able to estimate the groundspeed in pure head and tail winds, even in sections where thebirds experienced neither pure head nor tail winds (Fig. 1F). Weregarded wind speed as one-half the difference between maxi-mum and minimum ground speed values, and wind direction asthe direction at the maximum ground speed of the fitted sinu-soidal curve of each sections (Fig. 1 C and F, and Materials andMethods). The 5-min interval of the analysis sections was aconsequence of the trade-off between the need for sufficientnumbers of data points to estimate wind while keeping hightemporal resolution. It is also comparable to 10-min to 1-h in-terval of in situ measurement from buoys and weather stations.The advantage of using these seabirds is the characteristic dy-namic soaring flight, which relies heavily on the energy extractedfrom wind (23, 24), indicating the strong influence of wind ve-locity to their ground speed. This flight strategy results in atortuous trajectory including more than 10 soaring cycles of10–20 s, providing sufficient variation of flight direction in ashort time period that enabled a successful fitting of a sinusoidalcurve (Fig. 1 C and F, andMaterials and Methods), even when thebird seemed to fly in a certain direction over a large scale (Fig. 1D and E). We avoided erroneous estimation of wind when therewas no relationship between ground speed and flight directiondue to ambiguous flight behavior, although it excluded onlyunder 3% of wind estimation (Materials and Methods).

Results and DiscussionBird-Based Winds Were Correlated with Satellite-Based WindMeasurements. To examine the accuracy of the bird-based windvelocities, we compared it with satellite-based wind velocitiesestimated by the QuikSCAT and Advanced Scatterometer(ASCAT) satellite scatterometers. Many of the bird-based wind

15

10

5

00 90 180 270 360

Gro

und

spee

d (m

s-1)

15

10

5

00 90 180 270 360

Gro

und

spee

d (m

s-1)

Flight direction (°)

500 m

500 m

20 m

20 m

C

F

B

ED

A

Wind speed (ms-1)

Wind direction (°)

Wind speed (ms-1)

Wind direction (°)

N = 300s. d. ± 60.6°

N = 300s. d. ± 24.8°

Fig. 1. (A) An example of a 5-min section of the flight path of a streaked shearwater. The red arrow indicates the estimated wind velocity. (B) Enlarged viewof a meandering path shown in A. (D) Another example of a 5-min section of a flight path of a streaked shearwater when the bird seemed to travel in acertain direction. (E) Enlarged view of D showing repeated zigzag movement from a soaring maneuver. (C and F) The relationship between flight directionand ground speed of the path section in A [estimated wind speed of fitted curve, 3.11 ms−1; upper confidence interval (CI), 3.13 ms−1; and lower CI, 3.10 ms−1;estimated wind direction of fitted curve, 278°; upper CI, 280°; and lower CI, 277°] and D (estimated wind speed of fitted curve, 4.20 ms−1; upper CI, 4.34 ms−1;and lower CI, 4.10 ms−1; estimated wind direction of fitted curve, 304°; upper CI, 310°; and lower CI, 298°), respectively. Angular SDs of the flight direction are(C) 60.6° and (F) 24.8°, respectively. The red curve is the fitted sinusoidal curve. Gray area represents the 95% CI of the fitted sinusoidal curve. Estimated windspeed and direction is indicated by black arrows.

9040 | www.pnas.org/cgi/doi/10.1073/pnas.1523853113 Yonehara et al.

Dow

nloa

ded

by g

uest

on

Oct

ober

5, 2

020

measurements were located between the swaths or time regionsof the satellite coverage resulting in a total of only 20 collocatedcomparable points between bird-based and satellite-based winds[streaked shearwaters, n = 9 (Fig. 2 A a–c and B a–f, and Fig. S1);Laysan albatrosses, n = 2 (Figs. S2 A a and b and S3); andwandering albatrosses, n = 9 (Figs. S2 B a–i and S4)]. Thegeneralized vector correlation coefficient (25) accounting forboth wind speed and direction showed a significant correlationbetween the two methods (ρ2v = 1.66, n = 20, P < 0.01; Materialsand Methods). This was in the range of vector correlations shownin validation of satellite scatterometers measured winds comparedwith in situ measurements of winds by meteorological buoys (26, 27)(ρ2v = 1.28–1.90).Comparison of wind direction between bird and satellite-based

estimates showed good agreement (Fig. 3B; angular correlation

coefficient R = 0.44, P < 0.01). Absolute difference betweenbird-based and satellite-based measurements of wind directionbecame larger in weak winds (Fig. S5), particularly in somepoints from shearwaters and Laysan albatross (Fig. 3B), as shownin comparison between wind direction derived by buoys andsatellites (26, 27). Deviation from the true wind direction iscrucial in strong winds but minor in weak winds. This issue isaddressed by calculating the vector correlation coefficient con-sidering both speeds and directions, which showed good agree-ment between the two methods. Scatterometers evaluate oceansurface wind velocities by measuring the ocean surface rough-ness. In coastal areas, accuracy of satellite wind measurementsdecreases because wave structure becomes complex and small-scale wind variation caused by topography is greatest. Validationof satellite-based winds show that mismatches between windsderived by satellites and buoys occur most often near the shore(26, 27). This limited accuracy and difficulty in capturing small-scale wind variation near the shore might also explain the devi-ation of wind direction between the two methods, especially inshearwaters flying in the coastal areas. On the other hand, winddirections estimated from flight paths of wandering albatrossesshowed the strongest agreement with satellite-based wind directions(Fig. 3B) because this species flew in regions of strong persistentwinds and far away from land (Fig. S2B).The bird-based wind speed was strongly correlated with the

satellite-based wind speed (Pearson’s R = 0.94, P < 0.01) but wasunderestimated (Fig. 3A), which has several possible explana-tions. First, satellite-based wind speed is extrapolated to a 10-mreference height, whereas the average flight height of studiedbirds is below 10 m: ∼2 m for shearwaters and 3–8 m for alba-trosses (28). This difference in height is suspected to be one ofthe cause of the underestimation of bird-based wind speed dueto the shear of wind speed decreasing near the ocean surface. Toevaluate the discrepancy of wind speed due to height difference,we used the logarithmic wind profile near the ocean surface (29),which showed that the corresponding flight height would belower than 1 m to satisfy the underestimation of bird-based windspeed estimates (Fig. S6). This indicates that the wind shear doesnot solely explain the discrepancy between bird and satellite-based wind speed estimates. Second, potential sources of errorthat might be related to the characteristics of soaring flight of theseabirds should be considered. The unique dynamic soaringflight pattern used by shearwaters and albatrosses not only zig-zags in the horizontal direction but also undulates in the verticaldirection (23, 24, 30). Therefore, the variation in ground speedincludes the decrease and increase of ground speed associatedwith the gain and loss of altitude. Altitude variation related tomaneuvering of soaring birds consists of ascending in head windsand descending in tail winds. This may cause the estimated windspeed to deviate from the true wind speed experienced by thebird. However, studies of dynamic soaring flight show that po-tential energy associated with flight altitude is much smaller thanthe kinetic energy associated with fluctuating ground speed, in-dicating that wind resistance and assistance dominates groundspeed fluctuation (23, 24). Another error can be caused by al-batrosses and shearwaters adjusting their air speed in relation tohead and tail winds, with air speed increasing in head winds(21, 28), because here we assumed that the bird flew in a constantair speed in each section. Although flapping effort, which in-creases with decreasing body size (31), might also affect windspeed estimation especially in relatively small-sized shearwaters,this is thought to have small effect because intermittent flappingof soaring seabirds is considered to keep air speed in a certainrange for sustainable flight (30, 32) and we assumed that there isno rapid increase or decrease in ground speed by flapping, exceptduring takeoffs and landings, which we excluded from our anal-ysis (Materials and Methods). The strong correlation betweenbird-based and satellite-based wind speeds (Fig. 3A) suggest that

50 km

A

a

bc

10 km

a10 km

b

10 km

c

43°N

42°

41°

40°

141° 142° 143° 144°E

1086420Wind speed (ms-1)

1086420Wind speed (ms-1)

a b f

e

c

d

c

b

a d

e

f

B43°N

42°

41°

40°

142° 143° 144°E

10 km 10 km

10 km 10 km

10 km 10 km

50 km

Sanr

iku

Hokkaido

Fig. 2. (A) Bird-based winds are mapped on the flight paths of seven streakedshearwaters released on August 29, 2014, and (B) six streaked shearwatersreleased on September 2, 2014. Colors on the flight paths indicate wind speed,and black bars indicate the direction in which the wind is blowing. Stars in-dicate the breeding site. Points where bird-based wind was compared withsatellite-based wind are shown in boxes. Blue lines are the bird’s flight paths,and black arrows are the bird-based wind with length of the arrow indicatingwind speed. Red arrow in the boxes represents the bird-based wind temporallyclosest to the satellite-based wind measurements shown in gray arrows.

Yonehara et al. PNAS | August 9, 2016 | vol. 113 | no. 32 | 9041

ENVIRONMEN

TAL

SCIENCE

S

Dow

nloa

ded

by g

uest

on

Oct

ober

5, 2

020

bird-based wind speeds could be converted to comparable valuesfor practical use. Further analysis of the complex dynamics of theflight of these birds can increase estimation accuracy, especiallyby recording flight height to determine the reference height ofthe estimated wind velocities.

Bird-Based Wind Covers Spatial and Temporal Observation Gaps.Extensive travel distance and prolonged flight duration of soar-ing seabirds enabled fine-scale resolution and wide range esti-mation of wind speed and direction covering temporal and spatialgaps between the remote-sensing measurements. A total of 1,664bird-based wind data points was obtained from 13 streakedshearwaters released simultaneously on August 29 (n = 7) andSeptember 2 (n = 6), 2014. Each of the bird-based wind velocityrepresents the wind experienced by the birds during 5-min flightsof ∼2- to 3-km distance traveled, higher resolution compared withmore than 12-h and 12.5-km resolution of satellite-based windestimations. Estimated wind speed from streaked shearwatertracks ranged from 0.4 to 11.2 ms−1 with average of 3.4 ± 1.6 ms−1.Data points were widespread in the ocean between Hokkaido and

Sanriku in north eastern Japan and were densely distributed es-pecially near the Sanriku coast (approximately <100 km fromland) (Fig. 2) because the birds frequently returned to their colonyat the coastal island to feed their chicks. Dense distribution ofbird-based winds in coastal areas covered a key region wheresatellite-based wind measurements are lacking due to topographiceffects (Fig. 2 and Fig. S7 B and C). Offshore winds estimated bylong-distance foraging trips (∼500 km) were relatively strong,whereas speeds of coastal winds estimated by short-distance for-aging trips (approximately <100 km) were weaker and changeddirection frequently (Fig. 2). The high temporal resolution of thebird-based winds detected the dynamic change in wind directionfrom northerly winds to southerly winds that occurred between0:00 and 12:00 UTC of September 3 with timing differingaccording to the location of the birds (Fig. S7). These changeswere not recorded by the scatterometer wind estimation at 0:00UTC and 11:00 UTC on September 3 and 4, respectively, becauseof their lack of fine temporal resolution.

Possibility of Using the Bird-Based Wind. With this study, we haveprovided an initial indication of the use of seabird flight tracks toestimate ocean surface wind speed and direction with fine-scaleresolution both spatially and temporally. Wind velocities can beestimated where satellite-based wind measurements are lacking,such as in coastal areas. Blending such coastal in situ wind dataobserved by buoys into model analyses are known to improve thefidelity between modeled and observed coastal current systems(7). This suggests that the bird-based wind estimates could beused to cover the satellite measurement gaps. In fact, geophysicaldata directly measured by instrumented animals or indirectlyextracted from animal movements that complement conventionalobservation gaps have been assimilated to geophysical modelanalyses, as shown in currents extracted from shearwaters drift data(19) that were assimilated to ocean surface circulation models (20)and temperature and salinity profile measured by instrumentedseals (11) used in Antarctic circumpolar circulation models (12).Wind stress and wind stress curl in coastal areas induce surfacecurrents and upwelling systems that often generate areas of highbiological productivity (6, 8). We can intensively monitor theseproductive areas using seabirds because seabirds tend to forage insuch areas (19). These data can effectively complement the datafrom drifting buoys, which are less effective in detecting such areas.In addition, wind measurement coverage and validation can beeasily expanded by increasing the number of instrumented birds.Although indirect measurement from flight paths of soaring sea-birds might include errors related to behavior and flight strategy ofthe birds, bird-based and satellite-based wind velocities werestrongly correlated, suggesting that these estimated wind velocitieshave sufficient accuracy for further practical use. It is necessary tocarefully address the estimation errors caused by flight behaviorand difference between species through validation and furtherimprove the accuracy by analysis of 3D flight trajectories.The wind information derived from flight paths can be used in

bird’s movement ecology. Wind is treated as an important ex-ternal factor in studies of large-scale seasonal migrations (33–35)and commuting flights (36–38) of seabirds. Although the mi-gration patterns of seabirds associated with the large-scale windsystems are well documented (33–35), finer-scale movements inrelation to variation of winds in scales of several minutes tohours are still unrevealed due to the lack of wind informationrelevant to the fine-scale movements of the seabirds. Using themethods outlined in this study, high-resolution wind informationexperienced by birds throughout their whole flight can be eval-uated to investigate bird flight behavior in relation to more fine-scale turbulence and variation of wind.Soaring seabirds are exposed to various physical environments

throughout their journeys over the ocean. Integrating measure-ments of atmospheric and oceanographic variables measured

Bird-based wind speed (ms-1)

Sate

llite-

base

d w

ind

spee

d (m

s-1)

0

5

10

15

01 5150

wandering albatrossLaysan albatrossstreaked shearwater

A

Bird-based wind direction (°)

Sate

llite-

base

d w

ind

dire

ctio

n (°

)

072 06309 0810

072063

009

081

wandering albatrossLaysan albatrossstreaked shearwater

B

Fig. 3. (A) The relationship between bird-based (Vbird) and satellite-based(Vsat) wind speeds (n = 20, Vsat = 1.80  Vbird   −   1.42, Pearson’s R = 0.94, P <0.01). Gray area represents the 99% CIs. (B) The relationship between bird-based and satellite-based wind directions (n = 20, angular correlation R =0.44, P < 0.01). Dashed line represents equal speeds or directions. Plot colorrepresents streaked shearwaters (red dots), Laysan albatrosses (blue dots),and wandering albatrosses (black dots).

9042 | www.pnas.org/cgi/doi/10.1073/pnas.1523853113 Yonehara et al.

Dow

nloa

ded

by g

uest

on

Oct

ober

5, 2

020

by seabird-borne sensors or derived by seabird movements(13–16, 19) (e.g., water temperature, air pressure, ocean sur-face current, wind) presents a unique platform to monitor thepelagic environment by using seabirds as a fast-moving “livingocean buoy.”

Materials and MethodsField Study. In this study, we used data from three species of Procellariiformes:streaked shearwater (mean body mass, 0.6 kg), Laysan albatross (3.1 kg), andwandering albatross (9.7 kg). GPS loggers used in the field studies were GiPSy-2(Technosmart) for streaked shearwaters and Laysan albatrosses, and GPL20(Little Leonardo) for wandering albatrosses. GiPSy-2 was powered by Li-SOCl2battery and wrapped by heat shrink tube for waterproofing. The approxi-mate mass of the loggers were 25 g (GiPSy-2) and 80 g (GPL20), which cor-responded to less than 5% of bird’s body mass. We attached GPS loggers tothe back of the birds with waterproof tape (Tesa) and retrieved them afterthe birds returned to their nests. GPS loggers were set to take one positionalfix every second.

We attached the GPS loggers onto eight and nine streaked shearwaterssimultaneously on August 29 and September 2, 2014, respectively, at theFunakoshi–Ohshima Island breeding colony (39° 24′N, 141° 59′E) in Japan.Seven and eight loggers were retrieved, respectively. One bird lost its in-strument before recapture, and another bird was not recaptured. Two of theretrieved loggers did not record enough data. The remaining seven (nos.1–7) and six (nos. 11–16) datasets were used in further analysis. The proceduresof the field study were approved by the Animal Experimental Committee ofThe University of Tokyo, and this work was conducted with permission fromthe Ministry of the Environment and Agency for Cultural Affairs, Japan.

The field study of Laysan albatross was conducted in February 2014 at theKa’ena Point, Oahu Island breeding colony (21° 34′N, 158° 16′W) in Hawaii.GPS loggers were attached to three birds and all were recaptured. Onelogger did not record enough data, so the remaining two datasets (nos. 1–2)were used in further analysis. The experiment was conducted under per-mission from the Hawaii Department of Land and Natural Resources and theUS Geological Survey Bird Banding Laboratory.

The field study of wandering albatross was conducted in March 2007 atPossession Island, Crozet archipelago (46° 25′S, 51° 44′E) in the South IndianOcean. GPS loggers were attached to six birds, and all were recaptured. Twologgers failed to record due to exposure to seawater, and the remainingfour datasets (nos. 1–4) were used in further analysis. The experiment wasconducted under permission from the ethics committee of the InstitutePolaire Paul Emile Victor, France.

Wind Velocity Estimation. Analyses were done using Igor Pro (WaveMetrics).The developed software “Ethographer” (39) was used to calculate groundvelocity of each position. The ground velocity is composed of ground speed(V) and flight direction (θ). Ground speed (V) of each bird was calculatedfrom the distance between consecutive GPS positional fixes divided by thetime taken (1 s) between them. Flight direction (θ) was calculated as theclockwise angle between north and the line connecting each consecutiveGPS points. The frequency distribution of ground speed was bimodal for allthree species (Fig. S8). The lower ground speed values were interpreted asresting on ocean surface or on land, whereas higher ground speed valueswere interpreted as flight (40, 41). Based upon the bimodal distribution inthe histogram of ground speed, the resting phase was defined as a periodwhen a bird moved less than 4 ms−1 for more than 10 s (Fig. S8). Otherperiods were defined as the flight phase. Ground speed and flight directionduring the flight phase longer than 10 min were used to estimate oceansurface winds. One minute after takeoff and 1 min before landing were notused in our analysis considering the effect of frequent flapping accompa-nied with takeoff and landing (31). Each flight phase was divided into seriesof 5-min sections. The number of sections were 1,685 for streaked shear-waters, 718 for Laysan albatrosses, and 744 for wandering albatrosses. Av-erage latitude and longitude of each section were calculated to representthe positions of estimated wind data for each section. The angular deviation(42) of each flight section was calculated to represent the variation offlight direction.

Referring to Shimatani et al. (17), a sinusoidal curve was fitted against therelationship of ground speed (V) and flight direction (θ) for each 5-minsection using the following equation:

V =Va +a  sin θ+b  cos θ,

where Va, a, and b are coefficients. We assumed that birds flew with aconstant air speed (Va) in each 5-min section. By assuming that the ground

speed of a bird is only affected by the strength of wind speed along flightdirection, the ground speed of a bird would be maximized in pure tail windsand equals the sum of air speed and wind speed, whereas it would beminimized in pure head winds and equals the wind speed subtracted fromair speed. Therefore, wind speed of each section (Vw) was calculated as one-half of the difference between the maximum (Vmax) and the minimum (Vmin)values of the fitted sinusoidal curve using the following equation (Fig. 1 Cand F):

Vw = ðVmax −VminÞ=2.

We determined wind blowing direction of each section as θ corresponding tothe direction at the maximum flight speed (Vmax) of the fitted sinusoidalcurve (Fig. 1 C and F).

As wind speed becomes weaker, the peak of the fitted sinusoidal curvebecomes ambiguous, resulting in an unreliable estimation of wind direction.To avoid ambiguous estimation of wind direction when wind speed is ex-tremely low, Akaike’s information criterion (AIC) was calculated for sinu-soidal curve fitting (AICsin) and line fitting with a fixed slope of zero (AICnull)in each section, assuming normal distribution around each fitting. Wind wasnot estimated when

AICsin >AICnull − 2.

This selection of fitting also avoided wind velocity estimation when thevariation of flight direction was small or the ground speed variation seemedrandom, not reflecting the effect of wind speed and direction. The number ofsections in which wind estimation was avoided was 21 (1.2%) for streakedshearwaters, 18 (2.5%) for Laysan albatrosses, and 5 (0.7%) for wanderingalbatrosses. Excluding these points resulted in 1,664 wind data obtained fromstreaked shearwaters, 700 from Laysan albatrosses, and 739 from wanderingalbatrosses.

Accuracy Test of the Bird-Based Wind. To examine the accuracy, we comparedthe bird-based wind speed and direction with wind estimated by satellite-borne scatterometers. The scatterometers transmit microwave pulses to theocean surface and measures the surface roughness from the backscatteredpulses. Wind speed and direction are estimated by relating the surfaceroughness to wind stress. The satellites orbit around the earth two times perday and estimate wind speed and direction in continuous swaths coveringlarge parts of the global ocean. The reference wind speed and direction datawere obtained from the SeaWinds microwave scatterometer instrumentflown on the QuikSCAT spacecraft (43) (QSCAT) and Advanced Scatterometerinstrument flown on the EUMETSAT MetOp-A satellite (44) (ASCAT) fromPhysical Oceanography Distributed Active Archive Center (PODAAC)(podaac.jpl.nasa.gov/). We downloaded wind speed and direction data fromOPeNDAP in PODAAC where data were gridded in 12.5 × 12.5-km resolutionand wind speed was in a 10-m reference height for both QSCAT and ASCAT.Wind speed and direction data from QSCAT was used for comparison with2007 dataset (wandering albatross) and ASCAT for 2014 dataset (streakedshearwater and Laysan albatross). Bird-based winds and satellite-basedwinds were collocated for comparison by choosing the nearest point bothtemporally and spatially. Temporal difference and spatial separation werelimited to 2 h and 10 km, within the range of previous studies validating orcomparing with satellite-based winds (26, 27), resulting in maximum dif-ference of 89 min and 8.6 km. In many cases, the location of the bird-basedwind data was between the swaths of satellite-based wind measurements,which caused large spatial separation. Twelve out of 20 compared pointswere from different single individual (all points from shearwaters are fromdifferent individuals) and a maximum of four points is obtained from thesame individual (wandering albatross no. 2). Comparison points from thesame individual were temporally separated by at least 12 h, which couldsufficiently be treated as temporally independent observations.

Bird-based wind velocities and satellite-based wind velocities were bothdecomposed to x and y components in Earth-oriented Cartesian coordinates:x increases along the eastward axis, and y increases along the northwardaxis. The generalized vector correlation coefficient (25) (ρ2v) was calculated toevaluate the degree of correlation between bird-based and satellite-basedwind velocities. This coefficient takes into account both wind speed anddirection and shows a value between 0 and 2, with 0 indicating no corre-lation and 2 indicating complete correlation between two, 2D vector series.The generalized vector correlation coefficient is independent from thescaling effect to vector datasets by either a constant magnitude or angularshift. Therefore, only the wind speeds or the directions of satellite-based andbird-based winds were further compared. Wind speeds were compared byapplying Passing–Bablok regression (45) using mcreg function in mcr package

Yonehara et al. PNAS | August 9, 2016 | vol. 113 | no. 32 | 9043

ENVIRONMEN

TAL

SCIENCE

S

Dow

nloa

ded

by g

uest

on

Oct

ober

5, 2

020

in R, version 3.0.0 (46) (Fig. 3). Although the satellite-based wind estimatesused as the reference data for comparison also include error, Passing–Bablokregression compares two different methods (bird-based and satellite-based)estimating the same parameter (wind speed) taking into account that both ofthe estimating methods have an error. Confidence interval (CI) (99%) for theregression slope was calculated (upper CI, 2.73; lower CI, 1.40). Wind directionsof satellite-based and bird-based winds were compared by the nonparametrictest for angular–angular correlation (42) to test whether there was a corre-lation between them.

ACKNOWLEDGMENTS. We thank Tatsuo Abe, Takanori Abe, Bungo Nishizawa,Miho Sakao, and Tatsuya Shiozaki for the logistic support in the field. We hadfruitful comments from Suguru Kameda about the use of satellite data. We thankAkira Kuwano-Yoshida, Hiroshi Niino, Martina S. Müller, and Patrick J. O. Millerfor assistance to improve our manuscripts. Field studies were financially sup-ported by the Tohoku Ecosystem-AssociatedMarine Science; Bio-Logging Science,The University of Tokyo; and Japanese Society for the Promotion of ScienceGrants 24241001 and 16J09325. The work in the South Indian Ocean wassupported financially and logistically by the French Polar Institute (IPEV:programme no. 109) and the French Southern Territories administration (TAAF).

1. Liu WT (2002) Progress in scatterometer application. J Oceanogr 58(1):121–136.2. Chelton DB, Schlax MG, Freilich MH, Milliff RF (2004) Satellite measurements reveal

persistent small-scale features in ocean winds. Science 303(5660):978–983.3. Chelton DB, Freilich MH, Sienkiewicz JM, Von Ahn JM (2006) On the use of QuikSCAT

scatterometer measurements of surface winds for marine weather prediction. MonWeather Rev 134(8):2055–2071.

4. Ebuchi N, Graber HC, Caruso MJ (2002) Evaluation of wind vectors observed byQuikSCAT/SeaWinds using ocean buoy data. J Atmos Ocean Technol 19(12):2049–2062.

5. Pickett MH, Tang W, Rosenfeld LK, Wash CH (2003) QuikSCAT satellite comparisonswith nearshore buoy wind data off the U.S. west coast. J Atmos Ocean Technol 20(12):1869–1879.

6. Rykaczewski RR, Checkley DM, Jr (2008) Influence of ocean winds on the pelagicecosystem in upwelling regions. Proc Natl Acad Sci USA 105(6):1965–1970.

7. He R, Liu Y, Weisberg RH (2004) Coastal ocean wind fields gauged against the per-formance of an ocean circulation model. Geophys Res Lett 31(14):L14303.

8. Albert A, Echevin V, Lévy M, Aumont O (2010) Impact of nearshore wind stress curl oncoastal circulation and primary productivity in the Peru upwelling system. J GeophysRes 115(C12):C12033.

9. Kawai Y, et al. (2015) Marine atmospheric boundary layer and low-level cloud re-sponses to the Kuroshio Extension front in the early summer of 2012: Three-vesselsimultaneous observations and numerical simulations. J Oceanogr 71(5):511–526.

10. Biuw M, et al. (2007) Variations in behavior and condition of a Southern Ocean toppredator in relation to in situ oceanographic conditions. Proc Natl Acad Sci USA104(34):13705–13710.

11. Charrassin J-B, et al. (2008) Southern Ocean frontal structure and sea-ice formationrates revealed by elephant seals. Proc Natl Acad Sci USA 105(33):11634–11639.

12. Roquet F, et al. (2013) Estimates of the Southern Ocean general circulation improvedby animal-borne instruments. Geophys Res Lett 40(23):6176–6180.

13. Durant J, et al. (2009) Pros and cons of using seabirds as ecological indicators. Clim Res39(2):115–129.

14. Charrassin J-B, Park Y-H, Le Maho Y, Bost C-A (2002) Penguins as oceanographersunravel hidden mechanisms of marine productivity. Ecol Lett 5(3):317–319.

15. Weimerskirch H, Wilson R, Guinet C, Koudil M (1995) Use of seabirds to monitor sea-surface temperatures and to validate satellite remote-sensing measurements in theSouthern Ocean. Mar Ecol Prog Ser 126(1-3):299–303.

16. Wilson R, et al. (2002) Remote-sensing systems and seabirds: Their use, abuse andpotential for measuring marine environmental variables. Mar Ecol Prog Ser 228:241–261.

17. Shimatani IK, Yoda K, Katsumata N, Sato K (2012) Toward the quantification of aconceptual framework for movement ecology using circular statistical modeling. PLoSOne 7(11):e50309.

18. Treep J, et al. (2015) Using high resolution GPS tracking data of bird flight for me-teorological observations. Bull Am Meteorol Soc 97(6):951–961.

19. Yoda K, Shiomi K, Sato K (2014) Foraging spots of streaked shearwaters in relation toocean surface currents as identified using their drift movements. Prog Oceanogr 122:54–64.

20. Miyazawa Y, et al. (2015) Assimilation of the seabird and ship drift data in the north-eastern sea of Japan into an operational ocean nowcast/forecast system. Sci Rep5:17672.

21. Spear LB, Ainley DG (1997) Flight speed of seabirds in relation to wind speed anddirection. Ibis (Lond 1859) 139(2):234–251.

22. Wakefield ED, et al. (2009) Wind field and sex constrain the flight speeds of central-place foraging albatrosses. Ecol Monogr 79(4):663–679.

23. Sachs G, et al. (2012) Flying at no mechanical energy cost: Disclosing the secret ofwandering albatrosses. PLoS One 7(9):e41449.

24. Sachs G, Traugott J, Nesterova AP, Bonadonna F (2013) Experimental verification ofdynamic soaring in albatrosses. J Exp Biol 216(Pt 22):4222–4232.

25. Crosby DS, Breaker LC, Gemmill WH (1993) A proposed definition for vector corre-lation in geophysics—theory and application. J Atmos Ocean Technol 10(3):355–367.

26. Freilich MH, Dunbar RS (1999) The accuracy of the NSCAT 1 vector winds: Comparisonswith National Data Buoy Center buoys. J Geophys Res 104(C5):11231–11246.

27. Adams J, Flora S (2009) Correlating seabird movements with ocean winds: Linkingsatellite telemetry with ocean scatterometry. Mar Biol 157(4):915–929.

28. Pennycuick CJ (1982) The flight of petrels and albatrosses (Procellariiformes), observedin South Georgia and its vicinity. Philos Trans R Soc B Biol Sci 300(1098):75–106.

29. Stull RB (2003) An Introduction to Boundary Layer Meteorology (Kluwer Academic,Dordrecht, The Netherlands).

30. Spivey RJ, Stansfield S, Bishop CM (2014) Analysing the intermittent flapping flight ofa Manx shearwater, Puffinus puffinus, and its sporadic use of a wave-meanderingwing-sailing flight strategy. Prog Oceanogr 125:62–73.

31. Sato K, et al. (2009) Scaling of soaring seabirds and implications for flight abilities ofgiant pterosaurs. PLoS One 4(4):e5400.

32. Pennycuick CJ (1987) Flight of auks (Alcidae) and other northern seabirds comparedwith southern Procellariiformes: Ornithodolite observations. J Exp Biol 128(1):335–347.

33. Liechti F (2006) Birds: Blowin’ by the wind? J Ornithol 147(2):202–211.34. Shaffer SA, et al. (2006) Migratory shearwaters integrate oceanic resources across the

Pacific Ocean in an endless summer. Proc Natl Acad Sci USA 103(34):12799–12802.35. Croxall JP, Silk JRD, Phillips RA, Afanasyev V, Briggs DR (2005) Global circumnaviga-

tions: Tracking year-round ranges of nonbreeding albatrosses. Science 307(5707):249–250.

36. Jouventin P, Weimerskirch H (1990) Satellite tracking of wandering albatrosses.Nature 343(6260):746–748.

37. Weimerskirch H, Guionnet T, Martin J, Shaffer SA, Costa DP (2000) Fast and fuel ef-ficient? Optimal use of wind by flying albatrosses. Proc Biol Sci 267(1455):1869–1874.

38. Weimerskirch H, Louzao M, de Grissac S, Delord K (2012) Changes in wind patternalter albatross distribution and life-history traits. Science 335(6065):211–214.

39. Sakamoto KQ, et al. (2009) Can ethograms be automatically generated using bodyacceleration data from free-ranging birds? PLoS One 4(4):e5379.

40. Shiomi K, Yoda K, Katsumata N, Sato K (2012) Temporal tuning of homeward flightsin seabirds. Anim Behav 83(2):355–359.

41. Weimerskirch H, et al. (2002) GPS tracking of foraging albatrosses. Science 295(5558):1259.

42. Zar HJ (2010) Biostatistical Analysis (Pearson Prentice Hall, Upper Saddle River, NJ),5th Ed.

43. Lungu T (2006) QuikSCAT Science Data Product User Manual: Overview and Geo-physical Data Products, Version 3.0 (Jet Propulsion Laboratory, California Institute ofTechnology, Pasadena, CA).

44. Verhoef A, Stoffelen A (2016) ASCAT Wind Product User Manual, Version 1.14 (Oceanand Sea Ice Satellite Application Facility, EUMETSAT, Darmstadt, Germany).

45. Passing H, Bablok W (1983) A new biometrical procedure for testing the equality ofmeasurements from two different analytical methods. Application of linear re-gression procedures for method comparison studies in clinical chemistry, Part I. J ClinChem Clin Biochem 21(11):709–720.

46. R Core Team (2013) R: A Language and Environment for Statistical Computing(R Foundation for Statistical Computing, Vienna). Available at www.R-project.org.Accessed November 29, 2015.

9044 | www.pnas.org/cgi/doi/10.1073/pnas.1523853113 Yonehara et al.

Dow

nloa

ded

by g

uest

on

Oct

ober

5, 2

020